PEMS Meets Boiler NOx CEMS Requirements
By Aaron M. Cheng, Callidus Technologies Inc., and Ed Blankenship, Ultramar Diamond Shamrock
Predictive Emissions Monitoring Systems (PEMS) Can Replace Continuous Emissions Monitoring
Systems (CEMS) to continuously monitor stack emissions and be in compliance with both federal and state pollution regulations. These regulations include the Compliance Assurance Monitoring Rule and New Source Performance Standards.
A PEMS is developed by modeling the process and emission patterns of a piece of combustion-based equipment. It is installed and executed real-time on a computer system that has access to the data historian of the unit to be monitored. The PEMS has two models: sensor validation and emissions prediction. During the PEMS operation, the sensor validation model checks all the emissions prediction model input sensor values from the distributed control system (DCS). It activates alarms for failed sensors and reconstructs their values from historical data before the data are fed to the emissions prediction model. This dual model execution ensures a continuous and accurate PEMS for emissions monitoring in compliance with government regulations.
The emissions to be monitored generally include nitrogen oxides (NOx) and oxygen (O2). The first commercial PEMS was installed on a 221 MMBtu/hr gas-fired boiler at Arkansas Eastman, near Batesville, Arkansas, in May 1993. To date, about 100 PEMS have been installed for various boilers, furnaces and gas turbines throughout the United States.
There are three principal reasons to select a PEMS:
1. The initial and long-term costs for a PEMS can be much lower than those of a CEMS. Usually, the initial cost for a PEMS ranges from one-half to one-third that of a CEMS.
2. A PEMS can be used to optimize plant production while minimizing plant emissions.
3. In certain plants, because of the harsh environment of the emission sample collection system, a PEMS can be more accurate and reliable than a CEMS.
Ultramar Diamond Shamrock selected a PEMS to monitor NOx emissions for the West Plant Boiler 54-F-1, at Three Rivers, Texas. The company contracted with Callidus Technologies Inc., Tulsa, Okla., to develop a turn-key PEMS for 54-F-1 because of its combustion engineering expertise and extensive PEMS experience. Boiler 54-F-1 produces steam for process applications. It is rated at 212 MMBtu/hr maximum heat input and can generate 167,000 lb/hr steam at 470 psig and 710 F. Natural gas is the main fuel for the boiler.
Callidus engineers used their systematic methodology to develop an accurate and reliable PEMS for 54-F-1. The main steps were:
Design data collection plan and collect data;
Train emissions prediction and sensor validation models;
Install the PEMS; and
Conduct the Relative Accuracy Test Audit (RATA).
First, engineers designed a data gathering plan to collect process and emissions data from the boiler over a wide operating range. The test plan included the perturbation of several independent variables affecting NOx emissions, including excess O2 and fuel flow rate (along with its related steam production rate) at their low, medium and high conditions. An independent sampling and analysis company was contracted to collect emissions data for 54-F-1 in August 1997. Process data were recorded during the same period. Because of boiler operational changes, the team collected more data in October 1997. A total of 5,900 patterns (about 4 days) of data were collected during these two periods.
After the process and emissions data were collected and preprocessed, Callidus developed NOx emissions prediction models using a computer program for modeling. The initial model training used all of the available input sensors. Then a sensitivity analysis was performed to determine the dominant input sensors for emissions prediction. By using the top ranked inputs along with combustion engineering expertise, the model training process was repeated using a smaller set of inputs. Sixteen emissions prediction models were trained and tested. Project engineers then selected final NOx models that gave the best predictions and utilized a set of input sensors that represents the 54-F-1 operations from the combustion viewpoint. The final NOx emissions prediction model included six inputs, and the appropriate input sensor lower and upper bounds were set. The six sensor inputs from the emissions prediction model were then used for training the related sensor validation model. During real-time execution of the PEMS, the input variables must stay within the range defined by their lower and upper bounds in order for the model prediction results to be reliable.
The PEMS for boiler 54-F-1 was installed on a computer system running the UNIX operating system. Through a custom designed interface, the PEMS communicates with the DCS. It receives the input sensor data from the DCS, predicts NOx emissions and writes back the results to the DCS. These results are then archived in the data historian for reporting. The team-designed control console screens show the real-time NOx prediction results and the alarm status of the input sensors. Console operators can monitor the PEMS performance and take appropriate actions when sensors are drifting outside the model bounds, sensors are failing, or when the PEMS is down.
After the PEMS was installed and running on the computer system, appropriate RATAs were conducted for PEMS certification purposes.
Table 1 summarizes the basic statistics of the 1-minute PEMS NOx prediction results compared with the measured (or actual) values during model training. The statistics show that the minimum, maximum, mean and standard deviation of the PEMS predicted and measured NOx data are very similar. The average absolute relative error for predicted NOx is 2.37 percent. The average absolute deviation between the PEMS predicted and measured NOx is only 0.76 ppm. Figure 1 shows the 54-F-1 PEMS predicted vs. actual NOx (ppm) scatter plot with the lower and upper 20 percent error boundaries. The predicted NOx data are well within the allowable 20 percent error boundaries. These extremely accurate results gave a high level of confidence for the PEMS real-time prediction performance.
Callidus performed a sensitivity analysis of the effects of the six inputs on NOx emissions using the final NOx prediction model. Excess O2, rank #1, is the most important variable affecting NOx formation in boiler 54-F-1. The analysis indicates that NOx will increase with an increase in excess O2, when all the other five inputs are held constant.
In order to be certified, a PEMS for combustion equipment in Texas has to pass a PEMS RATA mandated by the Texas Natural Resource Conservation Commission. In general, a PEMS RATA has two parts: relative accuracy and statistical tests. A PEMS is fully certified if it can be shown to comply with the initial PEMS RATA requirements and three subsequent quarterly RATAs at low, medium and high levels of the key parameter affecting NOx. After four successful RATAs are conducted, the unit is required only to do semiannual RATAs (relative accuracy only) at normal load operations. However, if the latest RATA results indicate the PEMS predictions have a relative accuracy of less than or equal to 7.5 percent, then the next RATA can be conducted on an annual basis. The initial and three subsequent quarterly RATAs are only required for a unit if it is the first of its kind in the plant. All the other similar units with PEMS installed need to satisfy the initial RATA requirements and the following semiannual or annual RATAs. A typical PEMS RATA can be completed in one to two days.
In the relative accuracy part of the RATA, the PEMS predicted emissions are compared to the reference method (measured) data, and the results generally cannot exceed 20 percent relative accuracy (0 percent being perfect) for each of the three pre-selected levels. In the statistical part, an F-test, a t-test, and an r-value analysis are performed on the PEMS predicted results. F is calculated as the ratio of PEMS predictions variance to reference method data variance. The calculated F at each level cannot exceed the critical F-value determined at the 95 percent confidence level with (n-1) degrees of freedom, where n represents the number of pairs of predicted and measured data during the test. The F-test indicates how much the PEMS predictions variance differs from that of the reference method samples. The t-test determines whether bias adjustments are needed for the PEMS predictions vs. reference method values. The r-value analysis is a measure of how well the PEMS predictions correlate with the reference method data, with 1 being perfect and 0 indicating no correlation. The calculated r-value must be greater than or equal to 0.8. The purposes of the statistical tests are to verify the robustness of the PEMS and to ensure that it can predict accurate results under various operating conditions.
An independent sampling and analysis company performed the RATA tests for the Ultramar Diamond Shamrock Boiler 54-F-1 PEMS. The project team selected excess O2 as the key parameter to be perturbed at low, medium and high levels during the RATA because the PEMS sensitivity analysis indicated it was the most significant input affecting NOx emissions. The low, medium and high levels of excess O2 were assigned corresponding high, medium and low levels of steam production loads.
The PEMS passed its initial and first quarterly RATAs in January and May 1998, respectively. During each of these RATAs, the PEMS-predicted NOx (lb/MMBtu) data were tested against the reference method data. Table 2 shows the relative accuracy and statistical tests results of these two RATAs. In general, the PEMS exhibited relative accuracy results significantly better than the maximum allowable 20 percent limit. The first quarterly RATA shows that the predicted NOx results are all less than 7.5 percent under the three pre-selected levels. The F-values are far below the maximum limit of 1.86. The t-test indicates that a slight bias adjustment is needed for the NOx predictions. The r-values are significantly greater than 0.8.
If the PEMS sustains its current accurate performance, then long-term maintenance costs should be minimal as additional data collection and model retraining will not be needed. p
The authors would like to thank Callidus Technologies and Ultramar Diamond Shamrock for their support and permission to publish this paper, and Dr. Richard Martin and Dr. Kim Reyburn for their critical review of the paper.
Aaron M. Cheng was the principal engineer for developing PEMS models at Callidus Technologies, Tulsa, Okla. He has more than 18 years of computer modeling and simulation experience. He holds B.S., M.S. and Ph.D. degrees in engineering from the University of Oklahoma. Cheng is now with Aspen Technology Inc.
Ed Blankenship is the process controls superintendent at Ultramar Diamond Shamrock, Three Rivers Refinery, Three Rivers, Texas, with over 20 years of experience in process controls and instrumentation. He has a degree in instrument/electrical technology and a degree in computer systems.
Cheng, A.M., R.R. Martin and A.K. Reyburn, “Callidus` First Year Experience with Predictive Emissions Monitoring Systems (PEMS),” Paper presented at the Chemical Engineering Chemputers III Conference & Exhibition, Houston, Texas (Feb. 14-15, 1995).
Cheng, A.M. and G.F. Hagen, “An Accurate Predictive Emissions Monitoring System (PEMS) for an Ethylene Furnace,” Environmental Progress, Vol. 15, No. 1 (Spring 1996).
Cheng, A.M., R.R. Martin and A.K. Reyburn, “Field Experience with Predictive Emissions Monitoring Systems (PEMS),” Paper presented at the Forest Products Society Conference on Process and Business Control, Mobile Ala. (November 12, 1996).
Cheng, A.M., “Predictive Emissions Monitoring System (PEMS),” Presented at the ASME/A&WMA/DOE/EPA Emissions and Process Monitoring Workshop, Research Triangle Park, N.C. (December 11 – 12, 1996).
Cheng, A.M., “Monitoring Emissions of Various Types of Combustion Equipment by PEMS,” Paper presented at the Midwest Section Air & Waste Management Association 1997 Conference and Exhibition, Kansas City, Mo. (September 11, 1997).